Study Purpose and Population
In this cross-sectional study we investigated to determine if the use of dedicated property taxation to finance local public health agencies was associated with improved population health status. Taxation is used to redistribute resources. The ability to show benefits received from such redistributions is important for demonstrating the value and return on those taxpayer investments for public health services.
The population examined in this research represents 720 counties in eight states known as the Mississippi Delta Region (Alabama, Arkansas, Illinois, Kentucky, Louisiana, Mississippi, Missouri and Tennessee). Of the 720 counties in these states, 240 share a national designation as a Delta county because of common characteristics in population health status and socioeconomic conditions.
Data Collection
We collected county level data on the utilization of dedicated property tax levies to fund local public health agencies (i.e., services and facilities) in 720 counties of the eight states in the Mississippi Delta Region. Based on the U.S. 2000 Census, total population for the region was estimated to be 42 million. This study covered the 3-year period 2003-2005.
An initial step was to categorize the 720 counties as those that did or did not levy a dedicated tax for public health services. If a local government levied a tax dedicated for the local public health agency during the study period 2003-2005, we defined that county in this research as a county with a public health tax. There were a few instances where a county did not levy a tax for public health but a large city in that county did levy a tax dedicated for public health. If a city, with a population that represented more than 60% of the total county population, had a dedicated tax levy for the city public health agency, we considered that county as a county with a public health tax even if the county had not levied a tax for the local public health agency. For example, Green County and Buchanan County in Missouri were both considered counties with a public health tax because the City of Springfield in Green County and the City of St. Joseph in Buchanan County had city tax levies dedicated for the city public health agencies. The population in both of those cities represented over 60% of the total county population. Given the population size of those cities, we considered any potential impact of that tax to have influence on the majority of the county's population.
The legislative period for the property tax millage rate can vary from state to state, but is typically for multiple periods (i.e., 5 to 10 years). For any reason, if a tax levy dedicated for a county public health agency was not levied continuously during the study period, that county was considered a county without a public health tax. Based on these criteria, there were 338 counties with a public health tax and 382 counties without a public health tax in the study population. Fifty percent (n = 239) of the 480 non-Delta counties had a dedicated public health tax during the study period whereas only 41% (n = 99) of the 240 Delta counties had a dedicated public health tax. None of the counties in Arkansas, Mississippi, and Tennessee had a tax levy dedicated to public health agencies. Mississippi counties have authority to levy a dedicated tax for public health but Arkansas and Tennessee do not have such authority.
Outcome Measures
County-level data that are typically used as measures of community health status were collected for a set of health variables available from national, state, and local datasets. Invasive cancer incidence rates for each county were obtained from state cancer registries or the state health department. Other health outcome measures were obtained from the CDC Wonder website [
15]. We measured the health status of the counties using mortality rates for overall population, cardiovascular disease (CVD), cerebrovascular disease (stroke), heart disease, chronic lower respiratory disease (CLRD), diabetes, pneumonia/influenza, lung and bronchus cancer, all types of cancer and unintentional injury; incidence rates for lung and bronchus cancer, colorectal cancer, prostate cancer, female breast cancer and all types of cancer; and years of potential life lost rate before age 75 (YPLL75).
These health outcomes are a subset of the health outcomes studied by Studnicki et al. based on six categories: total population mortality, major disease mortality, cancer mortality and morbidity, avoidable hospitalizations, trauma/accidents mortality, and infectious diseases [
14]. We excluded the categories of avoidable hospitalizations and infectious diseases due to not having county level data for these outcomes or having small number of cases leading to unstable rates at the county level.
A county is considered to have an unreliable mortality rate for a disease when the number of deaths is less than 20 over the 3-year period studied. Additional file
1 shows the outcome variables included/excluded and the percent of counties with unreliable rates in the remaining four categories. Except for pneumonia/influenza and diabetes with around 43% percent of counties having unreliable data, the rest of the health outcomes we selected had less than 18% of counties with unreliable data. When a county had unreliable data for the health outcomes we selected, we used the Indirectly Standardized Mortality Rate for that county [
16].
The mortality rates, incidence rates, and rates for other outcome measures were calculated for the 3-year period 2003-2005 except for the Illinois cancer incidence rates which were for the 5-year period 2001-2005. Even if we hypothesize that the counties with a public health tax have better health outcomes compared to counties without a public health tax, there are other factors that can influence health outcomes differentially in different counties. Therefore, to adjust the difference in health outcomes between these two groups of counties due to these other factors, we used a regression analysis with a set of control variables. For control variables we selected demographic and economic variables: population size, percent net migration, percent population over 65, percent population under 18, percent non-white population, percent rural population, percent below poverty level, percent Medicaid eligible, number of physicians per 1,000 population, per capita income, unemployment rate, median household income, and whether the county is a Delta county or not. All of the control variables were expressed as annual averages during the study period except for percent rural population, which was based on year 2000 population. Population size was categorized into three groups: population less than 25,000, between 25,000 and 50,000, and over 50,000.
A preliminary analysis of the regression model indicated problems with multicollinearity due to highly correlated control variables. After removing these variables from the regression model, the control variables for the final model were population size, percent net migration, percent non-white population, percent population over 65, percent rural population, percent Medicaid eligible, number of physicians per 1,000 population, unemployment rate, per capita income, and whether the county is a Delta county or not.
Statistical Analysis
It is well known that health outcomes are highly correlated with per capita income [
17]. Plots of some outcome variables versus per capita income showed a quadratic relationship between health outcomes and per capita income. Similar trends have been seen in other studies. For example, a plot of life expectancy in the Organization for Economic Co-operation and Development (OECD) countries versus heath spending per capita shows a quadratic relationship [
18]. Therefore, we used the regression model,
where E(y) is the mean value of the health outcome variable, h = 1 or 0 depending on whether the county has a public health tax or not, c is the per capita income, f is a linear combination of the control variables and b
0, b
1, b
2, b
3, b
4, and b
5 denote the regression coefficients of the constant term, h, c, c
2, the interaction between h and c, and the interaction between h and c
2 respectively. The mean difference in health outcomes between counties with a public health tax and without a public health tax is then given by
which is a quadratic function of the per capita income. If none of the coefficients for quadratic terms in the regression were statistically significant, the variables corresponding to them were removed from the regression model and the mean difference in outcomes is linear in per capita income (E(y)|h = 1 - E(y)|h = 0 = b1 + b4c ). When linear terms also are not statistically significant, the mean difference in outcomes is the regression coefficient of h (E(y)|h = 1 - E(y)|h = 0 = b1).